1; RUN: llc -O0 -verify-machineinstrs -asm-verbose=0 -march=amdgcn < %s | FileCheck -check-prefix=SI -check-prefix=COMMON %s
2; RUN: llc -O0 -verify-machineinstrs -asm-verbose=0 -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=COMMON %s
3; XUN: llc -O0 -verify-machineinstrs -asm-verbose=0 -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=COMMON %s
4
5; SI hits an assertion at -O0, evergreen hits a not implemented unreachable.
6
7; COMMON-LABEL: {{^}}branch_true:
8define amdgpu_kernel void @branch_true(i8 addrspace(1)* nocapture %main, i32 %main_stride) #0 {
9entry:
10  br i1 true, label %for.end, label %for.body.lr.ph
11
12for.body.lr.ph:                                   ; preds = %entry
13  %add.ptr.sum = shl i32 %main_stride, 1
14  %add.ptr1.sum = add i32 %add.ptr.sum, %main_stride
15  %add.ptr4.sum = shl i32 %main_stride, 2
16  br label %for.body
17
18for.body:                                         ; preds = %for.body, %for.body.lr.ph
19  %main.addr.011 = phi i8 addrspace(1)* [ %main, %for.body.lr.ph ], [ %add.ptr6, %for.body ]
20  %0 = bitcast i8 addrspace(1)* %main.addr.011 to i32 addrspace(1)*
21  %1 = load i32, i32 addrspace(1)* %0, align 4
22  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %main_stride
23  %2 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
24  %3 = load i32, i32 addrspace(1)* %2, align 4
25  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr.sum
26  %4 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
27  %5 = load i32, i32 addrspace(1)* %4, align 4
28  %add.ptr2 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr1.sum
29  %6 = bitcast i8 addrspace(1)* %add.ptr2 to i32 addrspace(1)*
30  %7 = load i32, i32 addrspace(1)* %6, align 4
31  %add.ptr3 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr4.sum
32  %8 = bitcast i8 addrspace(1)* %add.ptr3 to i32 addrspace(1)*
33  %9 = load i32, i32 addrspace(1)* %8, align 4
34  %add.ptr6 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 undef
35  br i1 undef, label %for.end, label %for.body
36
37for.end:                                          ; preds = %for.body, %entry
38  ret void
39}
40
41; COMMON-LABEL: {{^}}branch_false:
42; SI: s_cbranch_vccnz
43; SI: s_cbranch_scc1
44; SI: s_endpgm
45define amdgpu_kernel void @branch_false(i8 addrspace(1)* nocapture %main, i32 %main_stride) #0 {
46entry:
47  br i1 false, label %for.end, label %for.body.lr.ph
48
49for.body.lr.ph:                                   ; preds = %entry
50  %add.ptr.sum = shl i32 %main_stride, 1
51  %add.ptr1.sum = add i32 %add.ptr.sum, %main_stride
52  %add.ptr4.sum = shl i32 %main_stride, 2
53  br label %for.body
54
55for.body:                                         ; preds = %for.body, %for.body.lr.ph
56  %main.addr.011 = phi i8 addrspace(1)* [ %main, %for.body.lr.ph ], [ %add.ptr6, %for.body ]
57  %0 = bitcast i8 addrspace(1)* %main.addr.011 to i32 addrspace(1)*
58  %1 = load i32, i32 addrspace(1)* %0, align 4
59  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %main_stride
60  %2 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
61  %3 = load i32, i32 addrspace(1)* %2, align 4
62  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr.sum
63  %4 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
64  %5 = load i32, i32 addrspace(1)* %4, align 4
65  %add.ptr2 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr1.sum
66  %6 = bitcast i8 addrspace(1)* %add.ptr2 to i32 addrspace(1)*
67  %7 = load i32, i32 addrspace(1)* %6, align 4
68  %add.ptr3 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr4.sum
69  %8 = bitcast i8 addrspace(1)* %add.ptr3 to i32 addrspace(1)*
70  %9 = load i32, i32 addrspace(1)* %8, align 4
71  %add.ptr6 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 undef
72  br i1 undef, label %for.end, label %for.body
73
74for.end:                                          ; preds = %for.body, %entry
75  ret void
76}
77
78; COMMON-LABEL: {{^}}branch_undef:
79; SI: s_cbranch_scc1
80; SI: s_cbranch_scc1
81; SI: s_endpgm
82define amdgpu_kernel void @branch_undef(i8 addrspace(1)* nocapture %main, i32 %main_stride) #0 {
83entry:
84  br i1 undef, label %for.end, label %for.body.lr.ph
85
86for.body.lr.ph:                                   ; preds = %entry
87  %add.ptr.sum = shl i32 %main_stride, 1
88  %add.ptr1.sum = add i32 %add.ptr.sum, %main_stride
89  %add.ptr4.sum = shl i32 %main_stride, 2
90  br label %for.body
91
92for.body:                                         ; preds = %for.body, %for.body.lr.ph
93  %main.addr.011 = phi i8 addrspace(1)* [ %main, %for.body.lr.ph ], [ %add.ptr6, %for.body ]
94  %0 = bitcast i8 addrspace(1)* %main.addr.011 to i32 addrspace(1)*
95  %1 = load i32, i32 addrspace(1)* %0, align 4
96  %add.ptr = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %main_stride
97  %2 = bitcast i8 addrspace(1)* %add.ptr to i32 addrspace(1)*
98  %3 = load i32, i32 addrspace(1)* %2, align 4
99  %add.ptr1 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr.sum
100  %4 = bitcast i8 addrspace(1)* %add.ptr1 to i32 addrspace(1)*
101  %5 = load i32, i32 addrspace(1)* %4, align 4
102  %add.ptr2 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr1.sum
103  %6 = bitcast i8 addrspace(1)* %add.ptr2 to i32 addrspace(1)*
104  %7 = load i32, i32 addrspace(1)* %6, align 4
105  %add.ptr3 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 %add.ptr4.sum
106  %8 = bitcast i8 addrspace(1)* %add.ptr3 to i32 addrspace(1)*
107  %9 = load i32, i32 addrspace(1)* %8, align 4
108  %add.ptr6 = getelementptr inbounds i8, i8 addrspace(1)* %main.addr.011, i32 undef
109  br i1 undef, label %for.end, label %for.body
110
111for.end:                                          ; preds = %for.body, %entry
112  ret void
113}
114
115attributes #0 = { nounwind }
116